CHAPTER 1 Elementary Arithmetic Sabine KOPPELBERG Freie Universitit Berlin Contents Introduction... : 1. Examples and arithmetic of Boolean algebras... 1.1. Definitions and notation 1.2. Algebras of sets. . 1.3. Lindenbaum-Tarski algebras. 1.4. The duality principle 1.5. Arithmetic of Boolean algebras. Connection with lattices. 1.6. Connection with Boolean rings. 1.7. Infinite operations .. 1.8. Boolean algebras of projections . . 1.9. Regular open algebras. : Exercises ...... 2. Atoms, ultrafilters and Stone’s theorem. . 2.1. Atoms. : 2.2. Ultrafiters and Stone's theorem. 2.3. Arithmetic revisited.......... 2.4. The Rasiowa-Sikorski lemma Exercises : 3. Relativization and disjointness ... 3.1. Relative algebras and pairwise disjoint families . 3.2. Attainment of cellularity: The Erdés-Tarski theorem. . 3.3. Disjoint refinements: the Balcar-Vojt48 theorem. Exercises .. = eerste HANDBOOK OF BOOLEAN ALGEBRAS Edited by J.D. Monk, with R. Bonnet © Elsevier Science Publishers BV., 1989Introduction Boolean algebras are defined by a list of algebraic axioms; this chapter deals with arithmetical laws derivable from the axioms and with notions arising in a particularly simple way from the arithmetic of Boolean algebras: atoms, relative algebras, and disjointness. We give, in Section 1, a somewhat lengthy list of axioms for Boolean algebras. There are many other, in particular shorter, axiom systems for Boolean algebras; a lot of work on these has been done but is outside the scope of this book. Our axiom system will, however, appear completely natural to a reader familiar with lattice theory: the axioms simply state that a Boolean algebra is a distributive and complemented lattice. We derive a certain amount of additional algebraic laws and give examples of Boolean algebras. The most sophisticated of these come from logic, functional analysis and topology: Lindenbaum-Tarski algebras, Boolean algebras of projections in Banach spaces, and regular open algebras of topological spaces. In Section 2 we use the laws established so far to prove Stone’s representation theorem, the most important fact in the structure theory of Boolean algebras: every Boolean algebra is isomorphic to an algebra of sets. This gives enough insight to dispense once and for all with proofs of further arithmetical laws since it implies that an equation holds in every Boolean algebra iff it holds in every algebra of sets. In Section 3 we prove elementary results on disjointness and two non-trivial theorems on disjoint families: the Balcar—Vojta theorem on disjoint refinements and the Erdés—Tarski theorem on the existence of large disjoint families, the latter one being a standard tool in combinatorial questions on Boolean algebras. We freely introduce and use some basic algebraic notions such as subalgebras, homomorphisms, etc. as they occur naturally in the proofs. They will be studied in greater detail in Chapter 2. 1. Examples and arithmetic of Boolean algebras This section presents a study of the arithmetic of Boolean algebras dad a variety of examples, the most important ones arising in set theory, logic, and topology: algebras of sets, Lindenbaum-Tarski algebras, and regular open algebras. 1.1. Definitions and notation 1.1. Derinimion. A Boolean algebra is a structure (A,+,°,—,0,1) with two binary operations + and -, a unary operation —, and two distinguished elements 0 and 1 such that for all x, y and z in A, (associativity) (B1) x+(y+z)=(e+y)+z, (B1') x-(y+z) =(x-y)-z,8 S. Korretpers / ELEMENTARY ARITHMETIC [ow 1 (commutativity) (B2) x+y=ytx, (B2') x-y=yrx, (absorption) (B3) x+(x-y)=x, (B3') x-(x+y)=x, (distributivity) (B4) x-(y+z)=(x-y)+(x-z), — (B4’) x +(y-2) =(e+y)-(@ +2), (complementation) (BS) x + (—x)=1, (BS') x-(—x) =0. Two standard examples of Boolean algebras, algebras of sets and Lindenbaum Tarski algebras, arise in set theory and logic. The operations +, - and — of a Boolean algebra are therefore often written as U, M, — or v, A, 7 and called union, intersection, complement or disjunction, conjunction, negation or, as in lattice theory, join, meet, complement. We follow Boole’s original notation which is frequently used in modern texts on axiomatic set theory and think about +, -, — as being spelled as sum, product, complement. It should be pointed out that the distinguished elements 0 and 1 of a Boolean algebra are not assumed to be distinct - see Example 1 The roles of + and - respectively of 0 and 1 in the above axiom system are completely symmetrical, a fact more thoroughly expressed in the duality principle 1.13. Nevertheless, to save notation we shall henceforth adopt the familiar convention that multiplication binds stronger than addition and omit parentheses around products whenever possible. We also agree that — binds stronger than both + and - and write, for example, —x + y- —z for (—x) + y-(—z), etc. By the first two couples of axioms, we tacitly omit parentheses and permute summands (respectively factors) in sums (respectively products). The structure (A, +,:,—,0,1) is usually identified with its underlying set A. This gives rise to the following definition. 1.2. Derinmion. A Boolean algebra (A,+,-,—,0,1) is finite (countable, of cardinality x, . . .) if its underlying set A is finite (countable, of cardinality x, . ..). ‘There is a perfectly natural notion of homomorphism between Boolean alge- bras which makes the class of all Boolean algebras, together with all homomor- phisms between them, into a category. When dealing with different Boolean algebras (A, + 45-4, —4s04+14) and (B, +,°p,—p, 0p, 1p), we drop the sub- scripts on +, -, etc. if no confusion arises. 1.3. DEFINITION. A homomorphism from a Boolean algebra A into a Boolean algebra B is a map f: A-> B such that f(0)=0, — f(1)=1, and for all x, y in A, fet y=fO)+fO), fle y)=f)- f(y) 5 f(x) = —f@) - fis an isomorphism from A onto B if fis a bijective homomorphism from A ontosi] EXAMPLES AND ARITHMETIC OF BOOLEAN ALGEBRAS 9 B. A and B are isomorphic (A= B) if there exists an isomorphism from A onto B. 1.2. Algebras of sets ‘The examples given in the following list are simple but quite basic and will often be referred to. Two of these, power set algebras and interval algebras, have very special properties not shared by every Boolean algebra. For example, power set algebras are complete but interval algebras of infinite linear orders are not, as explained in the subsection on infinite operations. On the other hand, algebras of sets provide the most general example of Boolean algebras: every Boolean algebra is, by Stone’s representation theorem 2.1, isomorphic to an algebra of sets, more precisely, by Theorem 7.8, to the clopen algebra of some topological space. 1.4, Exampue (power set algebras). Let X be any set and P(X) its power set. The structure (P(X), U, 9, —, 9, X), with —a the complement X\a of a with respect to X, is a Boolean algebra — the axioms (B1) through (BS') simply state elementary laws of set theory. P(X) is called the power set algebra of X. 1.5. EXaMPLe (the trivial Boolean algebra). For X the empty set, P(X) reduces to the Boolean algebra A=(P(0),...,04,1,) with 0,=1,, the trivial or one- element Boolean algebra. Obviously, any two Boolean algebras with exactly one element are isomorphic. 1.6. Exampre (the two-element Boolean algebra). For a singleton X = {x}, P(X) reduces to {0,1}, where 0=§ and 1= X. This algebra is called the nvo-element algebra and, following a convention in set theory, denoted by 2. Its operations are given by the table below. ioe) ei eet 0 0 0 1 01 0 1 Lie) 0 0 11 1 0 If 0 and 1 are identified with the truth values “false” and “true”, then the Boolean operations on 2 represent the logical operations of disjunction, conjunc- tion and negation. It follows from 1.18 and 1.21 below that any Boolean algebra with exactly two elements is isomorphic to 2.10 S. Kopreisers / ELEMENTARY ARITHMETIC [on 1 1.7. Derinimion. A structure (A,+4,*4,—4,04,14) is a subalgebra of the Boolean algebra (B,+5,°5,—9505, 1p) if ACB, 0,=0,, 14=1, and the operations + ,, "4, —4 are the restrictions of +5, °,, —, to A. Again we drop the subscripts whenever possible. By identification of Boolean algebras with their underlying sets, a subalgebra of B is simply a subset A of B containing 0, and 1, and closed under the operations of B. A is then a Boolean algebra in its own right since the axioms are valid for arbitrary elements of B and, a fortiori, of A. 1.8. Dermnmion. A subalgebra of a power set algebra P(X) is called an algebra of subsets of X or an algebra of sets over X. A Boolean algebra is an algebra of sets if it is an algebra of sets over X, for some set X. Note that A being an algebra of sets over X not only requires that the elements of A are subsets of X but also that the operations of A are the set-theoretical ones inherited from P(X) and that § and X are contained in A. 1.9. EXaMPLe (finite-cofinite algebras). Let X by any set. A subset a of X is called cofinite in X or, for fixed X, simple cofinite, if X\a is finite. Let A={aCX:a finite or cofinite} . A is then an algebra of sets over X, the finite—cofinite algebra on X. To check that aUb and aN bare in A for a, b in A, note that aU b is finite for a, b finite and cofinite otherwise; aM b € A follows from de Morgan’s law, aNb=X\(X\a)U(X\d)), since A is closed under — and U. If X has infinite cardinality x, then so has the finite—cofinite algebra on X, since X has exactly « finite subsets. Thus, every infinite cardinal is the cardinality of a Boolean algebra. A non-negative integer k, however, is the cardinality of a Boolean algebra iff k = 2" for some n € @, as follows from Corollary 2.8. 1.10. Exampte (clopen algebras). Let X be a topological space. A subset of X is clopen if it is both closed and open. The set Clop X of clopen subsets of X is an algebra of sets over X, the clopen algebra of X. 1.11. Exampte (interval algebras). Let L be a linearly ordered set with first element 0,. Extend the linear order of L to L U {~}, where © is an element not contained in L, by letting x<- for x © L. For x, y€ L U {~}, the set [x, y)={zEL:xsz(@ey) and modus ponens that a ~ y. Let [a] be the equivalence class of a with respect to ~ and put B(T) = {[a]: «a formula of L}. By defining [a]+[#]=[ev 6], [2] -[B]=[«»B], al fhe}, 1= [a> a6], 0=[a, A Tax] , where a, is an arbitrary formula, we make B(T) into a Boolean algebra, the algebra of (equivalence classes of) formulas with respect to T. In fact, the operations of B(T) are well-defined since Flaca’) s(BoB')> (av B)o@'v Bp’), etc.; the associative law (B1) holds in B(T) since Fav(Bvy)e(avB)vy, etc. For every formula a, [2] =1 iff T+ a. Thus, Tis syntactically consistent (i.e. not every formula is derivable from T) iff B(T) is not the trivial Boolean algebra. There are several naturally defined subalgebras of B(T), e.g. if we are deal- ing with first order logic and x,,...,2, is a fixed list of individual variables, then A=({[a]: all free variables of a are among x,,...,X,} is a subalgebra. In particular, the subalgebra {{a]: a a sentence of L) is called the Lindenbaum-Tarski algebra of T.81) EXAMPLES AND ARITHMETIC OF BOOLEAN ALGEBRAS 1B 1.4. The duality principle Before embarking on the arithmetic of Boolean algebras, we state a general principle which saves a good deal of computational work. We do not bother to formulate and prove it in great detail but think that even a reader inexperienced in logic will understand it. For every statement ¢ on Boolean algebras, let the dual statement d¢ of ¢ be obtained by systematically exchanging the symbols +, « and 0, 1 in . Obviously dd¢ = ¢ and each of the axioms (B;) in Definition 1.1 is the dual of (B,). It is this self-duality of the axiom system which gives rise to the following theorem. 1.13, THeoREM (duality principle). If a statement holds in every Boolean algebra, then so does its dual. Proor. Let ¢ hold in every Boolean algebra and let B= (B, +, -, —,0,1) be any Boolean algebra; we show that d@ holds in B. Consider the dual structure 4B =(B,:, +, —, 1,0) obtained from B by exchanging the operations + and - and the elements 0 and 1. An arbitrary statement y holds in dB iff dy holds in ddB = B. In particular by self-duality of the axiom system in Definition 1.1, dB is a Boolean algebra. So ¢ holds in dB and d¢ holds in B. Ol 1.5. Arithmetic of Boolean algebras. Connection with lattices We now prove a numbér of laws governing the elementary arithmetic of Boolean algebras. They will be used in the future without specific reference. All laws are assumed to be assertions about arbitrary elements x, y, z... of a Boolean algebra. By the duality principle, we shall prove only one of two dual statements in the following. 1.14, Lemma. (a) (idempotence) x + x =x and x-x=x. (b) xt y=yiffx-yax. Proor. (a) xtx=xtx-(xtx)_ by (B3’) ae by (B3). (b) Ifx+y=y, then, by (B3’), xry=xe(xty)=x, and the converse implication follows by duality. O4 S. Kopretsero / ELEMENTARY ARITHMETIC [cw 1 We take for granted the notion of a partial order or partially ordered set (P, =), ie. a set P with a binary relation = which is reflexive, transitive, and antisymmet- ric (if xSy and y Z,) is infinite, and there is some n such that both n and n+1 are in N. Then y,V a bounded self-adjoint transformation; it gives a representation f= J aB(aa) of f as the integral of the function id with respect to a spectral measure E. Here a spectral measure is a homomorphism from the Boolean algebra Bor R of Example 1.30 into a Boolean algebra A of projections in V which satisfies an additional requirement of countable completeness, similar to the definition 1.35 below. In the finite-dimensional case considered above, E would be the map defined by au)=[Zertsisn, eu} for each Borel set u in R. It is natural to require, for A a Boolean algebra of projections in a Banach space V, that the infinite operations of A, whenever defined, respect the topologi- cal structure of V. In view of (4), we define: 1.35. Derinimion. A Boolean algebra of projections in a Banach space is a complete (a-complete) algebra of projections if for each M C A (each countable MCA), £4 M and II* M exist and im(II* a2) = ime: ee My, im(24 m) = the topological closure of the linear subspace generated by Ufim e: eM). It can be shown that A is a complete Boolean algebra of projections iff, for each subset M of A, II* M (£4 M) is the limit, in the strong operator topology, of the net {114M': M’ C M finite} (of the net {Z“ M': M'C M finite}, respectively). Let us consider two simple examples of Boolean algebras of projections. For the Hilbert spacesi] EXAMPLES AND ARITHMETIC OF BOOLEAN ALGEBRAS 25 a : 2 H= {on nce! rma =} and uC o, define ¢,: H—> H by eu(X) =X° Xs (pointwise multiplication), where x,: @—> {0, 1} is the characteristic function of u. Then e, is a projection of H. A={e,:uCo} is a Boolean algebra of projections in H which is isomorphic to P(w) and hence complete as a Boolean algebra. It is not difficult to check that A is also a complete Boolean algebra in the sense of Definition 1.35. On the other hand, let J be an infinite set, V the Banach space of all bounded functions from / into R with the sup-norm; again for uC I, define e,: V->V by e,(x)=x-X,. Then A={e,:ucl, being isomorphic to P(I), is complete as a Boolean algebra but not complete in the sense of 1.35. To see this, let uCJ be the union of a strictly increasing sequence (u()),<, of subsets of u. If x denotes the map from / to R with constant value 1, then ¢,(x) =x, is not the limit of the sequence eyq)(x) = Xu) since lleu(x) — eun(2)I| = 1 for every n; thus (€,¢)),cu does not converge to ¢, in the strong operator topology. 1.9. Regular open algebras We give a standard example of a complete Boolean algebra, the regular open algebra of a topological space. For X a topological space and a C X, we denote by int a the interior and by cla the closure of a in X. 1.36. Derinrtion. Let X be a topological space. For aC X, ra=intcla is the regularization of a. uC X is regular open if ru =u. RO(X) = {uC X: u regular open} is the regular open algebra of X. ‘The name regular open algebra is justified by the following theorem where, for notational convenience, a Boolean algebra is identified with its associated partial order.26 S. Korreuners / ELEMENTARY ARITHMETIC [oud 1.37. TuEorEM. RO(X) is a complete Boolean algebra under set-theoretical inclu- sion. The distinguished elements and the operations of RO(X) are given by 0-8, 1=X, ut+v=r(uUv), u-v=uno, —u= int(XWu) , = m=-(Um), Im=-(nm). Proor. We establish six elementary facts on regular open sets. Let u, v, w range over open subsets of X. (6) wCru since uC clu and u = int u Cint cl uru. (6) If wand v are regular open, then so is uNv, since wv Cx(wM v) by (5), while (uM v) Cru =u and r(u Av) Crv = v imply that (wv) Cuno. (7) ra is regular open for each aC X: we have ra Ccla, clra Ccla, so rra =intclra Cintcla=ra. But ra Crra holds by (5). (8) wis the least regular open set including u , since uCru€RO(X) by (5) and (7), and wCv€RO(X) implies ru Crv =v. (9) uAraCr(uMa) for each aC X: openness of u implies uMclaCcl(uNa), hence uAra=int wu Mintcla= int(w Mel a) Cint (wa) = (uN). (10) Hunv=6, then wre =6 since uC X\v, cu Cel(XWw) = Xw, duNv=§ and rwAv=9. By the same reasoning, ruN rv = 9. Now for a proof of the theorem, note that (RO(X), C) is a partial order with a least and a greatest element, since both the empty set and X are regular open. For any subset M of RO(X), U M is the least set including each m€ M and it is open, so by (8) r(U M) is the least upper bound of M in RO(X). To show that w=1(A M) is the greatest lower bound of M, observe that w € RO(X) by (7). Since M M Cm for me M, it follows that w Crm = m; so w is a lower bound of a v is another lower bound of M in RO(X), then vCM M and v= (1M) = We have thus proved that (RO(X), C) is a lattice in which every subset has both a least upper bound and a greatest lower bound. In particular,sl] EXAMPLES AND ARITHMETIC oF BOOLEAN ALGEBRAS a (RO(X), +,-) satisfies the lattice axioms (B1) through (B3') since by (6), u-v=uNv for u, v€RO(X). Also w-int(X\u) = for u€ RO(X) since u and int(X\u) are disjoint sets and u + int(X\u) = 1 since u U int(X\w) is a dense subset of X. We are left with the proof of the distributive laws, and by Lemma 1.17 we need only check (B4). So let u, v, w be regular open. By the lattice axioms, we have uCutv, so w-uCw-(utv). Similarly, w-v Cw-(u + v) which gives w-u+ w-vCw-(w+tv). Conversely, w-(utv)=wAr(uUv) Cr(wA(uUv)) by (9) =1((w Nu) U(wNv)) =weutwey. O Even a very special case of this last theorem gives, up to isomorphism, all complete Boolean algebras: every partial order (P,<) can be topologized in a trivial way by taking the sets {q@P:qx, x is strictly smaller than y) if xy but x#y. At ={x€ A00 in a Boolean algebra A and there is no atom below x, then there exists a strictly decreasing infinite sequence X=x>x,>x,>-++in A*. The interval algebra of the real line (Example 1.11) is atomless and so is the regular open algebra of the reals in their usual topology (Theorem 1.37). To see the latter, note that for s < rin R, the open interval (s, t) of R is regular open. Now if a is a positive element of RO(R), then there are u0 and, for all x and y in A,a0 since otherwise, both a =x and a= —x hold for every x in A. (©) implies (a): assume that b € A and 0 = b B between the Boolean algebras A and B is a monomorphism or an embedding of A into B if it is one-to-one. It is an epimorphism it it is onto. ‘The image f[A] of A under a homomorphism f: A—> B is a subalgebra of B; if f is an embedding, then A is isomorphic to f[A] via f. 2.6. Proposition. For every Boolean algebra A, the map from A into the power set algebra P(At A) defined by30 S. Koppetperc / ELEMENTARY ARITHMETIC few. 1 f(x) = {aE At A:asx} is a homomorphism. It is an embedding if A is atomic and an epimorphism if A is complete. Proor. Obviously, f(0) =) and f(1) = At A. By 2.4(b), f(-x) = {@E At At as—x} =At A\{aG At A:asx} =At AV(x), and f(x + y) = f(x) U f(y) follows similarly from 2.4(0). Also, f(xy) =f) Of) since asx-y iffa P(X) is such an embedding or, more generally, a homomorphism, then for any point x of X, the subset p={aE A: x€e(a)} of A has the following properties: 1ép, 0¢p, a-b€p iffa€p and bep, a+bep iffacporbep, —a€p iffagp. ‘The subsets of A with these properties are exactly the ultrafilters of A defined below, as follows from Proposition 2.15. Thus, if e embeds A into P(X), then the points of X give rise to ultrafilters of A. Conversely, Stone’s theorem is proved by taking the ultrafilters of A as the points of a set Ult A; Corollary 2.16 then says that Ult A is large enough to embed A into P(UIt A). 2.10. Demin. A filter in a Boolean algebra A is a subset p of A such that 1ep, if x€p, yA and x0. 2.13, Lemma. The filter generated by E in A is the least filter of A including E. It is proper iff E has the finite intersection property. O Filters can also be characterized as being subsets p of A such that 1€p and (1) forallx, yEA,x-yEp iffxep and yep: if p is a filter and x- y Ep, then x Ep and y Ep since x-y 2 is a homomorphism from A into the two-element Boolean algebra, the characteristic homomorphism. Proor. For any filter of A, each of the three properties implies properness. Let p be maximal; we prove that p is an ultrafilter. Let x€ A. Since p is proper, x and —x cannot both be in p. Suppose xp. By maximality of p, the filter generated by p U {x} contains 0, hence by 2.13, a-x=0 for some ap. Thus, a=—x and —x Ep. Every ultrafilter is prime: assuming xp and yp, we have to show that x+yp. This follows from —x Ep, —~y Ep and —x--y=—(x+y)Ep. Primeness implies maximality: we assume x & p and prove that p U {x} gener- ates the improper filter. Since 1 = x + —x is in p but x is not, —x € p. So the filter generated by p U {x} contains x-—x=0. An arbitrary subset p of A is, by (1), a proper filter iff y,(1) = 1, x,(0) =0 and Xp Preserves the operation -. It is, by the first part of our proposition, an ultrafilter iff x, also preserves the operations + and —. 022] ATOMS, ULTRAFILTERS AND STONE'S THEOREM 3 Because of the equivalence between ultrafilters, prime filters and the prime ideals defined in Section 5 and also for historical reasons, the following existence theorem for ultrafilters, or rather its consequence that every non-trivial Boolean algebra has an ultrafilter, is called the Boolean prime ideal theorem (BPI). 2.16. Proposrrion (Boolean prime ideal theorem). A subset of a Boolean algebra is included in an ultrafilter iff it has the finite intersection property. Proor. If ECA is included in an ultrafilter p of A, then by properness of p, p and hence E have the finite intersection property. Conversely, assume EC A has the finite intersection property, so the filter py generated by E is proper by 2.13. The set P of all proper filters of A including py is non-empty and partially ordered by inclusion, moreover each non-empty chain C in P has UC as an upper bound in P. By Zorn’s lemma, let p be a maximal element of P. Then p is a maximal filter and includes E; by 2.15, it is an ultrafilter. O 2.17. Corottary. An element a of a Boolean algebra is contained in an ultra- filter iff a>0. Proor. The set {a} has the finite intersection property iffa>0. O The proof given above of the Boolean prime ideal theorem uses Zorn’s lemma, an equivalent of the axiom of choice, and it is in fact shown in FEFERMAN [1965] that BPI is not derivable from the axiom system ZF of Zermelo-Fraenkel set theory without the axiom of choice. On the other hand, HatPerN and Levy [1971] show that in ZF, BPI is strictly weaker than the axiom of choice. Like the axiom of choice, BPI has several interesting equivalences. For example, it is shown in Kextey [1950] that the full axiom of choice is equivalent in ZF to Tychonoff’s theorem (the product space of a family of compact topological spaces is compact) for arbitrary spaces, but the restriction of Tychonoff’s theorem to Hausdorff spaces is equivalent to BPI (cf. Los and Ryti-Narpzews«r [1954]; Rusiw and Scorr [1954]). With the machinery of ultrafilters at our hands, we are ready to embed any Boolean algebra into a power set algebra. 2.18. Dermviion. For A a Boolean algebra, Ult A={pCA: p an ultrafilter of A} is the set of ultrafilters of A. The map s: A— P(UIt A) defined by s(x) = {p €UIt A: x Ep} is the Stone map of A.34 S. Korretserc / Evementary ARiHMEniC [cw 1 Proof of Stone’s representation theorem 2.1. It follows as in the proof of 2.6 that the Stone map s is a homomorphism from A into P(Ult A). For example, s(0) =6 since every ultrafilter is proper, s(xty)={pixt yep} = {pix p}U{p: yep} = s(x) Us(y), since ultrafilters are prime, etc. We prove that s is one-to-one: let x #y in A; without loss of generality, xy. Thus, x-—y >0; by 2.17 let p be an ultrafilter containing x-—y. Then x € p and y &p which gives p €s(x)\s(y). O The set Ult A of ultrafilters of A and the Stone embedding s of A will be analyzed more thoroughly in Section 7: Theorem 7.8, the topological version of Stone’s theorem, states that Ult A can be topologized in such a way that s is an isomorphism from A onto the clopen algebra of Ult A. 2.3. Arithmetic revisited Stone’s theorem makes Boolean algebras look quite simple but, apart from giving an intuitive picture of how they look like, it does by no means trivialize every problem. For example, none of the combinatorial proofs in the following section would be really easier for algebras of sets than for arbitrary Boolean algebras. However, there is one aspect of their theory which is simplified by Stone’s theorem, or rather the method of ultrafilters developed for its proof — elementary arithmetic. Let us understand by an equation an expression of the form (Ga) = trees where f and 1’ are terms built up from the variables x,,...,x,, the constants 0 and 1 and the Boolean operation symbols +, -, and —. We say that the above equation holds in a Boolean algebra A if the values of ¢ and ’ coincide for every assignment of elements a,,...,4, of A to x,.-.,Xq- 2.19. Proposition. For every equation e, the following are equivalent: (a) ¢ holds in every Boolean algebra, (b) e holds in every power set algebra, (©) e holds in some non-trivial Boolean algebra, (d) e holds in the two-element Boolean algebra. Proor. Obviously, (a) implies (b) and (b) implies (). Also (c) implies (4) since the two-element algebra 2= {0, 1} is, in a canonical way, a subalgebra of every non-trivial Boolean algebra A; thus an equation holding in A holds, a fortiori, in 2.2] Atoms, ULTRAFILTERS AND Stone's THEOREM 35 Finally assume that (a) fails for an equation e of the form ¢(x,...x,)= (x, ...x,); We prove that (d) fails. There is a Boolean algebra A and a,,..., a, in A such that b = (a, ...a,) and b’=1'(a, ...a,) are distinct, say bb". But then, as in the proof of Stone’s theorem, there is an ultrafilter p of A such that bep and b' 2 be the characteristic function of p. Then 1= f(b) = flay. - aq) = fai) - - - flan) since f is a homomorphism from A into 2. Similarly, O= PC flay). « f@,)) 5 thus e does not hold in the two-element Boolean algebra. O The validity of a particular equation may thus be decided either by (b) above or, using the table in Example 1.6, i.e. the well-known truth-table method, by (d). We shall, in the rest of this text, not prove elementary equations holding in every Boolean algebra but leave it to the reader to convince himself by either method. Also, inequalities of the form #(x, ...x,) B be a homomorphism of Boolean algebras and MCA such that £4 M(II* M) exists. f preserves 2“ M (respectively II* M) if £* f[M] exists (IL” f[M] exists) and f(E4m) = 5" py (respectively f(II* M) =I? f[M]). Let p be an ultrafilter of A and MC A such that 5 M (II M) exists. p preserves 5 M if 5 M € p implies that m € p for some m€M. p preserves II M if MC p implies that II ME p. Thus, an ultrafilter p of A preserves 3 M (II M) iff its characteristic homomor- phism y,: A—>2 does. And by the infinitary version 1.33 of de Morgan’s laws, p preserves 0 M (respectively Il M) iff it preserves II {—m: m € M) (respectively 3 {-m: m€ M}). 2.21. THeorem (Rasiowa-Sikorski lemma). Let, in a non-trivial Boolean algebra A, S and P be at most countable families of subsets of A such that M exists for each M in S and IN exists for each N in P. Then there is an ultrafilter of A preserving © M for each M in S and IN for each N in P. Proor. By the remarks preceding this theorem, it suffices to find an ultrafilter preserving I M for each M in S. If S is empty, then any ultrafilter of A will do — note that A has at least one ultrafilter by the Boolean prime ideal theorem, being non-trivial. Thus, let S={M,:n€o)} by an enumeration of S. We construct by induction a decreasing sequence 1=a,24,2a,2--- in A* = A\{0} such that, for each n Eo, (2,) @yuy° M,=0 or a,,,0 has been constructed, we find a,,, as follows. If a,-£ M, =0, we let @,,,=4,. Otherwise by the distributive law 1.33(b), 00. So by (2,), a,,, =m for some m€ M, which implies m € p.92] Atoms, ULTRAFILTERS AND STONE'S THEOREM 37 One might naturally try to remove the countability restriction on the sets S and P in the Rasiowa-Sikorski lemma. This is partially done in the following statement; the “countable chain condition” referred to is defined in Section 3. Marmn’s axiom. Let, in a non-trivial Boolean algebra A satisfying the countable chain condition, S be a family of subsets of A such that |S|<2" and £ M exists for each M in S. Then there is an ultrafilter of A preserving 2 M for each M in S. Martin’s axiom is consistent with, but not provable from, the axioms of ZFC set theory. It is a consequence of the continuum hypothesis CH (2° = @,) since under CH it simply reduces to the Rasiowa-Sikorski lemma proved above in ZFC. Several consequences of CH can be also proved from Martin’s axiom; cf. MARTIN and Sotovay [1970] for a discussion of these topics. The analogue of Martin’s axiom to families § of cardinality at least 2°, however, contradicts the axioms of ZEC, see Exercise 3 in Section 3. And Exercise 7 in Section 4 shows that there is a Boolean algebra A not satisfying the countable chain condition and a family S of subsets of A such that |S] = @, and no ultrafilter of A preserves © M for each M in S. Exercises 1. In a Boolean algebra A, let MCA such that 5M exists. The Stone homomorphism s: A> P(Ult A) preserves E M iff £ M=5 M, for some finite subset M, of M. Similarly, let M and N be subsets of A such that 1 s[{M]C U s[N]. Then there are finite subsets M, of M and N, of N such that 1 s[M,]C U s[N]- 2. Let X be a compact Hausdorff space. Then for every ultrafilter p of RO(X), there is a unique point of X lying in M {clu: wep). 3. Let, in the regular open algebra RO((0, 1]) of the real unit interval, S be the set S={(a, b): a, b rational and 02= {false, true} under which every formula in T is true. Hint. Fix an ultrafilter p of the algebra B(T) and put h(v) = 1 iff [v] € p. Then for every formula a in L, @ is true under h iff [a] € p. 6. (Completeness theorem for countable languages in first order logic). Let L be a countable language for first order logic and T a consistent theory in L. Then T has a countable model. Hint. Fix, by the Rasiowa-Sikorski lemma, an ultrafilter p of B(T) preserving = M, for each L-formula (xx, ...x,); the set M, = {(tx, ...x,):¢a term in L} has been considered in Exercise 5 of Section 1. (a) Prove that t iff[t=]ep defines an equivalence relation = on the set of L-terms; let A be the set of equivalence classes t/= (b) Prove that there is a unique L-structure A with A as its underlying set such that, for each atomic L-formula a(x, ...x,,) and arbitrary terms t,,...,t,, () AR alt /= (©) Show that (+) holds for every L-formula a. It follows that A is a model of 7 -5t,/=] ifffa(t,...4,)Ep- 3. Relativization and disjointness We describe a simple construction of new Boolean algebras from old ones, the relative algebras. Relativization is a useful device since certain aspects of Boolean algebras, e.g. cardinal invariants, are sometimes easier to investigate for suitably chosen relative algebras of a Boolean algebra than for the algebra itself. The main topic of this section is the study of disjoint families and partitions of unity in a Boolean algebra, in particular of the cardinality functions of cellularity (respectively saturation) connected with these notions. It is not difficult to show that if X is a partition of unity in a sufficiently complete Boolean algebra A, then A is isomorphic to the cartesian product of the relative algebras A | x, x © X; see Section 6. Thus, questions about A can often be reduced to questions about some relative algebras of A. The method of relativization is applied in proving the first of two non-trivial combinatorial results on pairwise disjoint families, the Erdés—Tarski theorem on attainment of cellularity and the Balcar—Vojta% theorem on disjoint refinements. ‘They will be used in Section 13 where they play a prominent role in the proof of the Balcar-Franék theorem.83] RELATIVIZATION AND DISJOINTNESS 39 3.1. Relative algebras and pairwise disjoint families 3.1. LEMMA AND DEFINITION. Let A be a Boolean algebra and a € A. Then the subset A ba={xEA:xAta, p,(x)=a-x is a homomorphism which is onto since p,(x)=x for x€ A [ a. The cartesian product set A x B of two Boolean algebras A and B can be made into a Boolean algebra by defining all operations componentwise; we shall study cartesian products of (arbitrarily many) Boolean algebras in greater detail in Section 6. The following lemma explains the name of “factor algebras” for relative algebras. 3.2, Lemma. For each a in A, A=(A } a)x(A } -a). Proor. Define functions g:A—>(A}a)x(A}—a) and h: (Af a)x (A } -a)> A by a(x) =(x-a,x-—-a), Aly, z)=yt+z. Then g is a homomorphism since g(x) = (p,(x), p-.(X)) and it is bijective since g and h are inverses of each other. 3.3. Drrinition. Let A be a Boolean algebra, x and y in A and XC A. x and y are disjoint if x- y =0. X is a pairwise disjoint family if 00 and x,+x,=0 for i#j. 3.4. Proposition. In every infinite Boolean algebra, there are an infinite pairwise disjoint family, a strictly increasing infinite sequence and a strictly decreasing infinite sequence. Proor. Let A be infinite; we first construct a strictly decreasing sequence (@,) ney in A such that, for every nw, A } a, is infinite. Let a, = 1. Assume that a, has been constructed such that A } a, is infinite; pick a€ A | a, such that 0 «,) or complete algebras will be deter- mined in Section 12: they are exactly those cardinals « satisfying « 3.5. Coroxzary. An infinite o-complete Boolean algebra has cardinality at least ae Proor. Let A be infinite and o-complete; by Proposition 3.4, fix a pairwise disjoint family {d,,: m € @}. We show that the function f: P(w)—> A defined by f(M)= Bd, is one-to-one. If M and N are distinct subsets of @, then there is, without loss of generality, some m€ M\N. Then d,, = f(M), hence d,,-f(M) = d,,>0, whereas, by 1.33(b), 4,°fN)= Zidyed,=0. 0 It is a consequence of Zorn’s lemma that every pairwise disjoint family can be extended to a maximal one.$3) RELATIVIZATION AND DISIOINTNESS 41 3.6. LEMMA. A pairwise disjoint family X is maximal iff 8 X= 1. Poor. For the non-trivial direction of the equivalence, assume that I X does not exist or is strictly smaller than 1. Thus, there is an upper bound b of X satisfying b<1, and the pairwise disjoint family XU{—b} shows that X is not maximal. 0 In a power set algebra P(M), the maximal pairwise disjoint families are exactly the partitions of M, This motivates the following definition. 3.7. Derinmion. A subset of a Boolean algebra A is a partition (a partition of A, @ partition of 1, a partition of unity) if it is a maximal pairwise disjoint family. 3.2. Attainment of cellularity: the Erdés—Tarski theorem Proposition 3.4 implies that every countably infinite Boolean algebra has a countably infinite pairwise disjoint family. Is it possible to find, in an arbitrary infinite Boolean algebra A, a pairwise disjoint family of cardinality |A|? A strong counterexample to this question is provided in Section 9 by free Boolean algebras: they can have any prescribed infinite cardinality but have only countable pairwise disjoint families. We describe, in this subsection, the possible sizes of pairwise disjoint families in a Boolean algebra by the cardinal invariant of cellularity, the first of several cardinal functions on Boolean algebras to be studied later. In particular the countable chain condition is a very strong assumption on a Boolean algebra. It is relevant to set theory where it arises through the subjects of forcing and Martin’s axiom; cf. the discussion following the Rasiowa-Sikorski lemma in Section 2. 3.8. Drrinmion. Let A be a Boolean algebra, « a cardinal. Then cA =sup{|X|: X a pairwise disjoint family in A} and sat A=min{y: m a cardinal, |X| yu for some cardinal y, then there is a pairwise disjoint family X of size w in A | a. This holds since y 0}. Since £ X=1 by Lemma 3.6, we find that Y= U, cx Y,. Now {x-y: yEY,} is a pairwise disjoint family in A} x, so |Y,|Scx0}. It suffices to construct a pairwise disjoint family X such that |X(a,)|=« for a0 and A has no atoms; consequently, no ultrafilter of A is principal. Hence, no ultrafilter of A is, in the sense of Definition 2.12, generated by finitely many elements. 3.15. Proposition. Assume A is a x*-complete Boolean algebra and x* =cx for each x in A*. Then no ultrafilter of A is generated by less than x* elements. Proor. Assume for contradiction that p is an ultrafilter of A generated by the set E={e,:a A by slightly modifying the map f: P(w)— A in the proof of 3.5. 2. A subset X of a Boolean algebra A is called a chain (respectively a well-ordered chain) in A if X is, under the partial order inherited from A, a linear order (respectively a well-ordering). (a) If A satisfies the countable chain condition, as defined in 3.8, then each well ordered chain in A is countable. The converse holds if A is o-complete. (b) In the finite-cofinite algebra A over an infinite set, every chain is countable but A does not necessarily satisfy the countable chain condition. (© Find an interval algebra which has an uncountable chain but satisfies the countable chain condition. 3. Conclude from Exercise 3 in Section 2 that even for Boolean algebras satisfying the countable chain condition, the hypothesis |S|<2° cannot be removed from the formulation of Martin’s axiom in Section 2 without contradict- ing ZFC. 4. In the interval algebra of the reals, the subset {(a, b): @ and b rational, a0 implies (a) >0. Show that each algebra admitting a strictly positive finitely additive measure satisfies the countable chain condition.